Feature or enhancement

Proposal:

Please correct me if I misunderstood the code.

RemoteUnwinder.get_stack_trace() only looks at the head interpreter

• it gets the address of interpreters.head from _PyRuntimeState from these lines of code

  cpython/Modules/_remote_debugging_module.c

  Lines 2089 to 2106 in c19db1d

  uintptr_t address_of_interpreter_state;

  int bytes_read = _Py_RemoteDebug_PagedReadRemoteMemory(

  &unwinder->handle,

  runtime_start_address + interpreter_state_list_head,

  sizeof(void*),

  &address_of_interpreter_state);

  if (bytes_read < 0) {

  set_exception_cause(unwinder, PyExc_RuntimeError, "Failed to read interpreter state address");

  return -1;

  }

  if (address_of_interpreter_state == 0) {

  PyErr_SetString(PyExc_RuntimeError, "No interpreter state found");

  set_exception_cause(unwinder, PyExc_RuntimeError, "Interpreter state is NULL");

  return -1;

  }

  *interpreter_state = address_of_interpreter_state;
• _PyInterpreterState for the head interpreter is copied in these lines

  cpython/Modules/_remote_debugging_module.c

  Lines 2691 to 2699 in c19db1d

  char interp_state_buffer[INTERP_STATE_BUFFER_SIZE];

  if (_Py_RemoteDebug_PagedReadRemoteMemory(

  &self->handle,

  self->interpreter_addr,

  INTERP_STATE_BUFFER_SIZE,

  interp_state_buffer) < 0) {

  set_exception_cause(self, PyExc_RuntimeError, "Failed to read interpreter state buffer");

  goto exit;

  }
• Then, only iterate over PyThreadStates in that interpreter, and doesn't iterate over next interpreters:

  cpython/Modules/_remote_debugging_module.c

  Lines 2760 to 2785 in c19db1d

  while (current_tstate != 0) {

  PyObject* frame_info = unwind_stack_for_thread(self, &current_tstate);

  if (!frame_info) {

  Py_CLEAR(result);

  set_exception_cause(self, PyExc_RuntimeError, "Failed to unwind stack for thread");

  goto exit;

  }

  if (PyList_Append(result, frame_info) == -1) {

  Py_DECREF(frame_info);

  Py_CLEAR(result);

  set_exception_cause(self, PyExc_RuntimeError, "Failed to append thread frame info");

  goto exit;

  }

  Py_DECREF(frame_info);

  // We are targeting a single tstate, break here

  if (self->tstate_addr) {

  break;

  }

  // If we're only processing the GIL holder, we're done after one iteration

  if (self->only_active_thread && gil_holder != NULL) {

  break;

  }

  }

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• gh-138385: Sample all interpreters in the tachyon profiler #138398